Integrating Statistical and Machine Learning Models for Predicting the Lifetime of High‐Luminosity AlGaInP Red LEDs Under Multi‐Stress Conditions

ABSTRACT This study explores the integration of statistical methods and machine‐learning models to predict the lifespan of high‐luminosity AlGaInP red light‐emitting diodes (LEDs) subjected to Accelerated Life Testing (ALT). The analysis involved seven different stress scenarios that varied in injected current, temperature, and relative humidity. Two lifetime distributions, Weibull and lognormal, were employed, alongside life‐stress relationships based on the Peck and Intel models. The evaluation of model performance revealed that the lognormal distribution consistently captures the degradation behavior of the LEDs. The Intel model showed a modestly better fit than the Peck model for the present dataset, as indicated by its slightly higher coefficient of determination and lower AIC and RMS values. While machine learning models trained solely on ALT data struggled to predict lifespan under normal operating conditions, training across the full operational range substantially reduced their prediction errors. Statistical correlation and regression analyses identified temperature as the primary factor influencing LED life, followed by humidity and drive current. This finding was corroborated by the calculated activation energy, which was half the bandgap of AlGaInP, suggesting that thermal stress activates mid‐gap traps and is a critical factor in degradation.